Selected work

Representative Papers

P2GS: Physical Prior-guided Gaussian Splatting for Photometrically Consistent Urban Reconstruction

May 16, 2026

This work addresses the photometric inconsistencies and artifacts in static backgrounds caused by varying camera exposures and dynamic illumination in real-world driving scenarios when using conventional 3D Gaussian splatting. To resolve this, we propose the first Gaussian splatting framework that integrates a physical imaging model, enabling joint disentanglement of a view-invariant linear HDR radiance field, per-view exposure scales, and tone-mapping functions from only LDR images. By incorporating relative exposure consistency constraints and HDR-domain radiance regularization, our method achieves cross-view photometrically consistent reconstruction without requiring HDR supervision. Evaluated on both real and synthetic driving datasets, the approach attains state-of-the-art LDR reconstruction quality while significantly improving photometric consistency, reliability of exposure normalization, and physical plausibility of recovered illumination.

0 citationsRead paper

FlashSAC: Fast and Stable Off-Policy Reinforcement Learning for High-Dimensional Robot Control

Apr 06, 2026

This work addresses the slow convergence and training instability commonly observed in off-policy reinforcement learning for high-dimensional robotic control, which often stem from error accumulation in the critic. The authors propose an enhanced Soft Actor-Critic framework that, for the first time, incorporates scaling principles from supervised learning into off-policy reinforcement learning. By increasing model capacity and data throughput, reducing gradient update frequency, and explicitly constraining the norms of weights, features, and gradients to suppress error propagation, the method achieves significantly better performance than PPO and strong off-policy baselines across more than 60 tasks in 10 simulated environments. Notably, it excels in high-dimensional dexterous manipulation tasks and reduces sim-to-real transfer training time for humanoid robots from hours to minutes.

0 citationsRead paper

SWE-Bench++: A Framework for the Scalable Generation of Software Engineering Benchmarks from Open-Source Repositories

Dec 19, 2025

Existing repository-level software engineering benchmarks (e.g., SWE-bench) rely on manual annotation, static datasets, and are limited to Python defect repair, lacking multilingual support and scalability. This work introduces the first scalable, multilingual (11 languages) repository-level evaluation benchmark, automatically constructed from real GitHub Pull Requests to yield executable defect repair and feature development tasks. We propose a novel PR-driven, four-stage automation pipeline: programmatic collection, containerized environment synthesis, test oracle extraction, and quality assurance—augmented by prompt-guided failure trajectory synthesis for model training. The benchmark comprises 11,133 instances across 3,971 repositories. On a 1,782-sample subset, Claude-3.5-Sonnet achieves 36.20% pass@10. Fine-tuning on this benchmark significantly improves performance on SWE-bench Multilingual, demonstrating its efficacy for training and evaluation of multilingual code intelligence models.

0 citationsRead paper

Read or Ignore? A Unified Benchmark for Typographic-Attack Robustness and Text Recognition in Vision-Language Models

Dec 10, 2025

Current LVLM evaluation and defense strategies prioritize ignoring image-text to enhance robustness, yet this compromises essential multimodal reasoning—such as jointly recognizing persons and interpreting traffic signs—in real-world scenarios. Method: We propose “Read-or-Ignore Visual Question Answering (RIO-VQA)”, a novel task formalizing context-adaptive decisions on whether to read text within images. To support it, we introduce RIO-Bench—the first benchmark featuring counterfactual paired data and a “read-or-ignore” selective text utilization paradigm—and design the first data-driven adaptive defense framework integrating counterfactual image generation, selective attention, and robust VQA modeling. Contribution/Results: Our analysis exposes a fundamental misalignment between existing evaluation protocols and practical multimodal reasoning requirements. Experiments show that state-of-the-art LVLMs and defenses fail to balance robustness and text understanding; our approach significantly improves RIO-VQA accuracy, establishing a new pathway toward reliable multimodal reasoning.

0 citationsRead paper

Text-Printed Image: Bridging the Image-Text Modality Gap for Text-centric Training of Large Vision-Language Models

Dec 03, 2025

To address the high acquisition cost, privacy constraints, and domain-specific data scarcity associated with real images in vision-language model (VLM) training, this paper proposes the Text-Printed Image (TPI) paradigm: text descriptions are directly rendered as monochromatic textual images on a pure white background, enabling zero-cost construction of semantically faithful synthetic image-text pairs. TPI bypasses generative models entirely, relying solely on lightweight text rendering, and integrates seamlessly into existing VLM training pipelines. Coupled with large language models to generate diverse, high-quality captions, TPI facilitates purely text-driven vision-language pretraining and data augmentation. Comprehensive evaluation across four state-of-the-art VLMs and seven benchmarks demonstrates that TPI significantly outperforms diffusion-based, text-centric training approaches—achieving superior generalization, practicality, and scalability while preserving semantic fidelity and reducing computational and ethical overhead.

0 citationsRead paper

Recent publications

Latest Papers

P2GS: Physical Prior-guided Gaussian Splatting for Photometrically Consistent Urban Reconstruction

May 16, 2026

This work addresses the photometric inconsistencies and artifacts in static backgrounds caused by varying camera exposures and dynamic illumination in real-world driving scenarios when using conventional 3D Gaussian splatting. To resolve this, we propose the first Gaussian splatting framework that integrates a physical imaging model, enabling joint disentanglement of a view-invariant linear HDR radiance field, per-view exposure scales, and tone-mapping functions from only LDR images. By incorporating relative exposure consistency constraints and HDR-domain radiance regularization, our method achieves cross-view photometrically consistent reconstruction without requiring HDR supervision. Evaluated on both real and synthetic driving datasets, the approach attains state-of-the-art LDR reconstruction quality while significantly improving photometric consistency, reliability of exposure normalization, and physical plausibility of recovered illumination.

0 citationsRead paper

FlashSAC: Fast and Stable Off-Policy Reinforcement Learning for High-Dimensional Robot Control

Apr 06, 2026

This work addresses the slow convergence and training instability commonly observed in off-policy reinforcement learning for high-dimensional robotic control, which often stem from error accumulation in the critic. The authors propose an enhanced Soft Actor-Critic framework that, for the first time, incorporates scaling principles from supervised learning into off-policy reinforcement learning. By increasing model capacity and data throughput, reducing gradient update frequency, and explicitly constraining the norms of weights, features, and gradients to suppress error propagation, the method achieves significantly better performance than PPO and strong off-policy baselines across more than 60 tasks in 10 simulated environments. Notably, it excels in high-dimensional dexterous manipulation tasks and reduces sim-to-real transfer training time for humanoid robots from hours to minutes.

0 citationsRead paper

SWE-Bench++: A Framework for the Scalable Generation of Software Engineering Benchmarks from Open-Source Repositories

Dec 19, 2025

Existing repository-level software engineering benchmarks (e.g., SWE-bench) rely on manual annotation, static datasets, and are limited to Python defect repair, lacking multilingual support and scalability. This work introduces the first scalable, multilingual (11 languages) repository-level evaluation benchmark, automatically constructed from real GitHub Pull Requests to yield executable defect repair and feature development tasks. We propose a novel PR-driven, four-stage automation pipeline: programmatic collection, containerized environment synthesis, test oracle extraction, and quality assurance—augmented by prompt-guided failure trajectory synthesis for model training. The benchmark comprises 11,133 instances across 3,971 repositories. On a 1,782-sample subset, Claude-3.5-Sonnet achieves 36.20% pass@10. Fine-tuning on this benchmark significantly improves performance on SWE-bench Multilingual, demonstrating its efficacy for training and evaluation of multilingual code intelligence models.

0 citationsRead paper

Read or Ignore? A Unified Benchmark for Typographic-Attack Robustness and Text Recognition in Vision-Language Models

Dec 10, 2025

Current LVLM evaluation and defense strategies prioritize ignoring image-text to enhance robustness, yet this compromises essential multimodal reasoning—such as jointly recognizing persons and interpreting traffic signs—in real-world scenarios. Method: We propose “Read-or-Ignore Visual Question Answering (RIO-VQA)”, a novel task formalizing context-adaptive decisions on whether to read text within images. To support it, we introduce RIO-Bench—the first benchmark featuring counterfactual paired data and a “read-or-ignore” selective text utilization paradigm—and design the first data-driven adaptive defense framework integrating counterfactual image generation, selective attention, and robust VQA modeling. Contribution/Results: Our analysis exposes a fundamental misalignment between existing evaluation protocols and practical multimodal reasoning requirements. Experiments show that state-of-the-art LVLMs and defenses fail to balance robustness and text understanding; our approach significantly improves RIO-VQA accuracy, establishing a new pathway toward reliable multimodal reasoning.

0 citationsRead paper

Text-Printed Image: Bridging the Image-Text Modality Gap for Text-centric Training of Large Vision-Language Models

Dec 03, 2025

To address the high acquisition cost, privacy constraints, and domain-specific data scarcity associated with real images in vision-language model (VLM) training, this paper proposes the Text-Printed Image (TPI) paradigm: text descriptions are directly rendered as monochromatic textual images on a pure white background, enabling zero-cost construction of semantically faithful synthetic image-text pairs. TPI bypasses generative models entirely, relying solely on lightweight text rendering, and integrates seamlessly into existing VLM training pipelines. Coupled with large language models to generate diverse, high-quality captions, TPI facilitates purely text-driven vision-language pretraining and data augmentation. Comprehensive evaluation across four state-of-the-art VLMs and seven benchmarks demonstrates that TPI significantly outperforms diffusion-based, text-centric training approaches—achieving superior generalization, practicality, and scalability while preserving semantic fidelity and reducing computational and ethical overhead.

0 citationsRead paper